The present invention relates to a fault monitoring method for computer, and in particular to a fault monitoring method of an operating system (OS) and an application program (AP).
In general, an AP encounters a fault for several reasons and stops. Main causes are defects included in the AP itself and faults of an OS on which the AP is mounted. As for such an AP that its operation needs to be continued, the following method has been adopted. A fault is monitored. If a fault has been detected, then the operation of the AP is resumed from a state in which it was before the occurrence of the fault, and thereby recovery of the AP from the fault is attempted.
As one of fault monitoring methods for AP and OS, there is a method called watchdog. “Fault Tolerant Computer” written by Takashi NANYA, published by OHM-SHA says “The watchdog timer is a different process which is independent of a monitored process. The monitored process is so designed that a timer is reset at intervals of a fixed time (for example, in the range of several microseconds to several minutes) during the execution of the monitored process. If the timer is not reset until the time is up, some fault is considered to have occurred in the monitored process.”
In the case where a fault of an AP is monitored, an AP fault monitor having a watchdog receives a periodical alive message from the AP. If the alive message stops for a predetermined time, the AP fault monitor judges an AP fault to have occurred and restarts the AP. In the case where a fault of an OS is monitored, an OS fault monitor having a watchdog receives a periodical alive message from the OS. If the alive message disappears for a predetermined time, the OS fault monitor judges an OS fault to have occurred and restarts the OS.
Furthermore, for recovery of an AP from a fault, a technique called check-point is used. In the check-point technique, execution information of the AP is acquired from the monitored AP periodically and reserved, When a fault has occurred, the reserved execution information is taken out, and processing of the AP is resumed from a check-point.
As for an AP fault monitor using software, the AP fault monitor operates on the same OS as the monitored AP. In some cases, therefore, the AP fault monitor using software cannot cope with such an AI′ fault caused by the OS. Furthermore, there is known such an OS fault monitoring method that a watchdog is formed of dedicated hardware and the watchdog monitors a periodical alive message supplied from an OS. However, there is a problem that dedicated hardware must be prepared.
On the other hand, as for the check-point, if check-point information is preserved in a low speed memory, then it takes a considerable time to preserve the check-point information, and consequently the check-point repetition period becomes long. As a result, recovery from a fault must be conducted from a state of a considerably long time before the time of the fault occurrence. Furthermore, it is possible to provide a high speed non-volatile memory, such as a static RAM, apart from a volatile memory, such as a dynamic RAM, managed by the OS, and preserve check-point information in the high speed non-volatile memory. However, there is a problem that dedicated hardware must be prepared.